Youth with Special Educational Needs (SEN) face challenges in the employment market. Job interview skills represent a barrier due to anxiety and communication difficulties. Traditional human-led training programs suffer from scalability limitations, inconsistent delivery, and anxiety induction. This study presents JoBot, a cat-shaped social robot designed to provide engaging, adaptive job interview training in a low-stakes environment. We conduct an empirical study with 28 youth with SEN (ages 19-26) randomly assigned to robot-led (n=14) or human-led (n=14) training sessions. This study employs a mixed-methods approach that incorporates performance assessments, questionnaires, and qualitative interviews. Training efficacy in four interview competence domains is evaluated. We also examine psychological outcomes using Self-determination Theory. In general, the experimental group shows greater improvement (mean: +17.06%) than the control group (mean: +8.23%). For performance assessment, participants' verbal competence exhibits statistically significant enhancement (robot-led: +30.19%, p<.01; human-led: +16.96%, p=.13). For psychological outcomes, robot-led training manifests a statistically significant improvement in intrinsic motivation (robot-led: +7.69%, p<.05, d=.58; human-led: +2.86%, d=.33). Qualitative analysis reveals that most participants demonstrate motivational readiness. Some participants report that they increase confidence but reduce stress through repeated robot-led practice. These findings suggest that robot-led training creates a stress-free environment and fosters a non-judgmental mastery-learning experience for youth with SEN. This work provides empirical evidence that social robots are effective and scalable alternatives for job interview training among SEN populations. The result implies the importance of inclusive and assistive technology development in the future.